Experiments in Stable Diffusion

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • EthylTheAardvark
    Senior Member
    • Nov 2016
    • 169

    Experiments in Stable Diffusion

    Submitted for your consideration :-D

    My results are hit or miss, frequently way more miss than hit. I'm running the program locally on a Nividia 4090 card using Easy Diffusion in Windows. This is generated using the stable diffusion 2.1 model, which seems to prefer higher resolution images to lower ones. I'm experimenting with generating a subject using the replace syntax after some number of steps in the process. In this case the keywords [vulpera:transluscent red latex balloon:25] trying to get it to replace the generated model with a new material. Was pretty effective in this one case, although the failures can frequently be nightmare fuel.


    Having done both, so far I've found that working with an artist is much easier than working with an AI.
    Attached Files
  • aron_crow
    Senior Member
    • May 2020
    • 171

    #2
    Re: Experiments in Stable Diffusion

    Oh wow, I have been trying with online AI tools since a long time and its always a bummer. I once trained a model on Runway.AI and it did pretty decent but that being said it was far far from perfect, you would find better renditions on google images. I'm hoping for the days when this tech becomes better and we are also able to generate/train videos, because as it happens I have >12k videos & 100k looner images and I can only imagine the kind of content that will generate.

    By the way, how have you trained the network?

    Comment

    Working...